A forest carbon sink remote sensing monitoring method and system

Abstract

The present invention relates to the technical field of forest carbon sinks, and more specifically, to a forest carbon sink remote sensing monitoring method and system, which are used to solve the problem that the prior art can obtain a carbon sink prediction result by inputting a target area image into a preset carbon sink model, but cannot combine with a vegetation index to build a more comprehensive carbon storage prediction model, and the use of the preset model cannot effectively cope with the characteristics of a specific task, resulting in inaccurate prediction results. The present invention combines ground survey data and vegetation indices through a prediction model construction module to build a comprehensive carbon storage prediction model, and obtains a carbon storage prediction model through machine training and learning, which can effectively cope with the characteristics of a specific task, thereby improving the accuracy and flexibility of the prediction, and also generates a high-resolution carbon storage distribution map through an efficient machine learning algorithm, supports scientific decision-making and management, and provides an important tool for carbon storage research.

Description

Technical Field

[0001] The present invention relates to the technical field of forest carbon sinks, and more specifically, to a forest carbon sink remote sensing monitoring method and system. Background Art

[0002] Forest carbon sink refers to the absorption of carbon dioxide in the atmosphere by forest plants and its fixation in vegetation or soil, thereby reducing the concentration of the gas in the atmosphere. Forests are the largest carbon reservoir in terrestrial ecosystems and play a very important role in reducing the concentration of greenhouse gases in the atmosphere and mitigating global warming. Remote sensing monitoring of forest carbon sinks is of great significance. Through remote sensing technology, rapid and accurate monitoring of forest carbon sinks can be achieved, providing strong support for formulating scientific forestry policies and promoting forestry informatization.

[0003] The patent application with reference publication number CN115829118A discloses a forest carbon sink remote sensing monitoring method, device, equipment and storage medium, the method comprising: extracting multiple forest area images from the original remote sensing image of the target forest area taken; obtaining the climate characteristics of the target forest area through the historical weather library of the target forest area and extracting characteristic factors; performing radiation correction on the multiple forest area images according to the characteristic factors, and integrating the corrected multiple images to obtain the target forest area image; inputting the target forest area image into a preset carbon sink model to obtain a carbon sink amount prediction result. Since the present invention only needs to obtain remote sensing data and query historical library data, the degree of intelligence is improved, and the carbon sink amount is predicted by a preset model at the same time, without using a fixed calculation method such as a theoretical formula, accurate monitoring of forest carbon sinks is achieved, and monitoring efficiency is improved;

[0004] However, the existing technology obtains the target area image through radiation correction and integration of images, and the processing of remote sensing images is not in-depth and comprehensive enough, which reduces the accuracy of image data and is not conducive to the subsequent analysis and model building of image data; at the same time, the existing technology can obtain the carbon sink prediction result by inputting the target area image into a preset carbon sink model, and cannot build a more comprehensive carbon storage prediction model in combination with the vegetation index. Moreover, the use of the preset model cannot effectively cope with the characteristics of specific tasks, which will lead to inaccurate prediction results, and cannot evaluate the effectiveness of the carbon storage prediction model, and cannot provide a basis for the parameter optimization of the model, which reduces the accuracy and reliability of the model.

[0005] To this end, we propose a forest carbon sink remote sensing monitoring method and system to address the above issues. Summary of the invention

[0006] The purpose of the present invention is to provide a forest carbon sink remote sensing monitoring method and system, which solves the problem that the prior art obtains the target area image by performing radiation correction and integration on the image, the remote sensing image processing is not in-depth and comprehensive enough, the accuracy of the image data is reduced, and it is not conducive to the subsequent analysis and model construction of the image data; at the same time, the prior art can obtain the carbon sink prediction result by inputting the target area image into a preset carbon sink model, and it is impossible to construct a more comprehensive carbon storage prediction model in combination with the vegetation index, and the use of the preset model cannot effectively cope with the characteristics of a specific task, which will lead to inaccurate prediction results, and it is impossible to evaluate the effectiveness of the carbon storage prediction model, and it is impossible to provide a basis for the parameter optimization of the model, thereby reducing the accuracy and reliability of the model.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] A forest carbon sink remote sensing monitoring method comprises the following steps:

[0009] Step 1: Set monitoring content and select monitoring areas. Monitoring content includes carbon storage change monitoring and forest health status monitoring;

[0010] Step 2: Collect high-resolution remote sensing images covering the entire monitoring area, obtain multi-temporal images, and collect ground survey data. Multi-temporal images are a set of images of the same area obtained at different time points. Ground survey data include tree height, tree diameter at breast height, tree species composition and forest density;

[0011] Step 3: Perform geometric correction and radiation correction on the remote sensing images, stitch multiple images into a complete image, and crop them to the monitoring area;

[0012] Step 4: Calculate vegetation indices based on standardized images, analyze time series data, and assess forest growth and seasonal changes;

[0013] Step 5: Use ground survey data and vegetation index to build a carbon storage prediction model, apply machine learning methods to calculate the carbon storage of each pixel, and generate a comprehensive carbon storage distribution map;

[0014] Step 6: Compare the carbon storage predicted by the model with the field measurement data, evaluate the effectiveness of the carbon storage prediction model, and write a monitoring report.

[0015] As a preferred embodiment of the present invention, the specific process of setting the monitoring content and selecting the monitoring area is as follows:

[0016] The monitoring content is set as carbon storage change monitoring and forest health status monitoring. The carbon storage change monitoring method is as follows: set monitoring time intervals to capture changes in carbon storage, use a combination of remote sensing and ground survey methods, use vegetation indices and ground survey data to integrate and build a carbon storage prediction model;

[0017] The forest health monitoring methods are as follows: analysis of canopy changes, vegetation cover and vegetation index based on remote sensing images to assess forest health, regular ground surveys to record tree growth, pest and disease impacts and canopy thickness;

[0018] Before selecting the monitoring area, the specific objectives of the monitoring should be determined, which are: assessing the changes in forest carbon stocks and analyzing forest health, developing selection criteria for monitoring areas, and using geographic information systems for spatial analysis to map out appropriate monitoring areas;

[0019] The specific contents of the selection criteria are as follows: Select representative forest ecosystems;

[0020] Preference shall be given to primary forests or restored forest lands;

[0021] Select areas where carbon stock changes are significant;

[0022] The selected areas are easy to access and perform monitoring in practical operations.

[0023] As a preferred embodiment of the present invention, the specific steps of performing geometric correction and radiation correction on the remote sensing image are as follows:

[0024] Geometric correction: Select an image with precise geographic reference as the reference image, select n known ground control points from the correction image and the reference image, and mark the image coordinates of these control points (x m ,y m ) and the corresponding geographic coordinates (Xm, Ym), where m = 1, 2, ..., n, the image coordinates are converted to geographic coordinates using the affine transformation model. The conversion formula is as follows: , where (X', Y') is the corrected geographic coordinate, (x, y) is the pixel coordinate in the image to be corrected, ɑ, b, c and d are affine transformation parameters, tx and ty represent the translation in the X and Y directions respectively;

[0025] The objective function is constructed by minimizing the control point error:

[0026] , where S is the sum of squares of residuals between control points, (X m , Y m ) is the actual geographic coordinate, (X m ', Y m') is the corrected coordinate calculated by the transformation model;

[0027] The obtained affine transformation parameters are used for the image to be corrected, and the corrected image is generated by resampling;

[0028] Radiation correction: Convert the original spectral brightness value of the image into reflectance. The conversion formula is as follows:

[0029] , where R λ is the reflectivity at wavelength λ, L λ is the measured brightness at wavelength λ, θ is the solar altitude angle, E λ is the solar radiation intensity at wavelength λ;

[0030] The mean-standard deviation method was used for radiation compensation. The correction factor was calculated by comparing the standard image and applied to the corrected image.

[0031] As a preferred implementation of the present invention, the specific steps of stitching and cropping remote sensing images are as follows:

[0032] Image stitching: Select feature matching points in the overlapping area and use feature extraction algorithms to identify matching points;

[0033] Use the linear fusion method to stitch images together. The stitching formula is as follows:

[0034] M final (x,y)=αM 1 (x, y) + (1-α) M 2 (x, y),

[0035] Among them, M final (x, y) is the pixel value of the stitched image at coordinate (x, y), M 1 (x, y) and M 2 (x, y) is the pixel value of the two images at the coordinate (x, y), α is the balance factor, 0<α<1;

[0036] Use image processing tools to perform image stitching. The stitched images must have no obvious discontinuity in color and brightness. The stitched images are output as a new image file.

[0037] Image cropping: Use vector layers to define the boundaries of the required cropping area. The required analysis area must be completely included in it, and the stitched images are cropped along the boundaries of the cropping area.

[0038] As a preferred embodiment of the present invention, the specific process of evaluating the growth status and seasonal changes of the forest is as follows:

[0039] The acquired multi-temporal images are subjected to geometric correction, radiation correction, splicing, cropping and normalization to generate standardized images corresponding to each time period. The standardized images are saved and exported to obtain the reflectance NIR of the near-infrared band, the reflectance RED of the red band and the reflectance BLUE of the blue band. The enhanced vegetation index EVI of each time period is calculated by the following formula:

[0040] , where G is the gain factor, e 1 and e 2 is the adjustment factor and L is the background correction factor.

[0041] As a preferred embodiment of the present invention, the EVI value of each phase calculated by the formula is recorded and a time series data set is formed, missing value processing and outlier detection are performed on the data in the time series data set, an EVI time series graph is drawn, the moving average of the time series is calculated to smooth the data, and linear regression analysis is used to identify long-term trends;

[0042] The forest growth condition is judged by the EVI value: when the EVI value is greater than or equal to the preset forest health growth threshold, it indicates that the forest growth condition is good; when the EVI value is less than the preset forest health growth threshold, it indicates that the forest growth condition is poor;

[0043] Determine the start and end time when the EVI value is greater than the preset growth threshold, mark the time when the EVI value is first greater than the preset growth threshold as the start date of the growing season, and mark the time when the EVI value is last greater than the preset growth threshold as the end date of the growing season. The calculation formula for the growing season length GSL is as follows: GSL=end date-start date.

[0044] As a preferred embodiment of the present invention, the specific process of constructing the carbon reserve prediction model is as follows:

[0045] The ground survey data and the calculated enhanced vegetation index EVI are used as feature quantities related to carbon storage prediction to construct feature set A, which includes tree height, tree diameter at breast height, tree species composition, forest density and EVI value. Feature set A is converted into a corresponding set of feature vectors, which are used as the input of the machine learning model, and the carbon storage corresponding to each set of feature vectors is used as the output of the machine learning model. With carbon storage as the prediction target and the minimum sum of prediction errors of all training data as the training target, the machine learning model is trained until the sum of prediction errors reaches convergence and the training is stopped to obtain the carbon storage prediction model. The carbon storage prediction model is expressed as follows:

[0046] C=β 0 +β 1 H+β 2 DBH+β 3·D+β 4 EVI+β 5 ·D s +ε;

[0047] Where C is the predicted carbon storage, β 0 is the intercept, H is the tree height, DBH is the tree diameter at breast height, D is the forest density, and D s is the variable of tree species composition, after necessary coding, β 1、 β 2、 β 3、 β 4 and β 5 is the regression coefficient of each input feature, and ε is the random error term.

[0048] As a preferred embodiment of the present invention, the specific process of calculating the carbon storage of each pixel and generating a comprehensive carbon storage distribution map is as follows:

[0049] The tree height, tree diameter at breast height, forest density and EVI value data corresponding to each pixel are collected, and the collected data are converted into a corresponding set of feature vectors and input into the carbon storage prediction model. The carbon storage of each pixel is calculated through the carbon storage prediction model, and the calculated carbon storage of each pixel is mapped to the spatial coordinates to form a comprehensive carbon storage distribution map.

[0050] As a preferred embodiment of the present invention, the specific process of evaluating the effectiveness of the carbon reserve prediction model is as follows:

[0051] Obtain the field measurement data for analysis. The field measurement data include tree height, tree diameter at breast height, forest density, EVI value and measured carbon storage in the monitoring area. Obtain the predicted data of the carbon storage prediction model. The predicted data is the predicted carbon storage of the monitoring area calculated by the carbon storage prediction model. Perform missing value processing and standardization on the field measurement data and the predicted data. Calculate the mean absolute error MAE of the carbon storage prediction model by the following formula:

[0052] ;

[0053] in, It is Prediction values, It is The actual measured value, is the sample size;

[0054] Calculate the coefficient of determination R 2 To assess the variation explained by carbon stock prediction models:

[0055] ,in is the average value of the actual measured values;

[0056] If the MAE is less than or equal to the preset MAE threshold and R 2 The value is close to 1, indicating that the carbon storage prediction model is effective;

[0057] If the MAE is greater than the preset MAE threshold, it indicates that the carbon storage prediction model is invalid;

[0058] With the X-axis representing the carbon storage predicted by the carbon storage prediction model and the Y-axis representing the carbon storage measured on site, a carbon storage scatter plot is drawn, and a 45° reference line is drawn in the plot;

[0059] The results of the comparison process are recorded in a monitoring report, which includes data charts, statistical results and analysis conclusions, and the monitoring report is sent to the communication terminals of relevant personnel.

[0060] As a preferred embodiment of the present invention, a remote sensing monitoring system for forest carbon sinks includes a monitoring platform, a monitoring setting module, an image data collection module, an image processing module, a forest growth assessment module, a prediction model construction module and a model assessment module;

[0061] The monitoring setting module sets the monitoring content and selects the monitoring area;

[0062] The image data collection module acquires multi-temporal images and collects ground survey data;

[0063] The image processing module performs geometric correction, radiation correction, splicing and cropping on the remote sensing image;

[0064] The forest growth assessment module assesses the growth status and seasonal changes of the forest;

[0065] The prediction model building module is used to build a carbon storage prediction model and calculate the carbon storage of each pixel;

[0066] The model evaluation module evaluates the effectiveness of the carbon storage prediction model.

[0067] Compared with the prior art, the advantages of the present invention are:

[0068] (1) In the present invention, the image processing module is used to perform geometric correction, radiation correction, splicing and cropping on the remote sensing image, which can significantly improve the accuracy and reliability of the image, eliminate spatial distortion and spectral differences, enhance the availability of data, and facilitate the subsequent analysis of the image data and construction of the model;

[0069] (2) In the present invention, a comprehensive carbon storage prediction model is constructed by combining ground survey data and vegetation index through a prediction model construction module. The carbon storage prediction model can effectively cope with the characteristics of specific tasks through machine training and learning, thereby improving the accuracy and flexibility of prediction. In addition, a high-resolution carbon storage distribution map is generated through an efficient machine learning algorithm, which supports scientific decision-making and management and provides an important tool for carbon storage research.

[0070] (3) In the present invention, the effectiveness of the carbon storage prediction model is evaluated through the model evaluation module, which can provide a basis for the optimization of model parameters, improve the accuracy and reliability of the model, provide important support for environmental management and policy formulation, enhance public trust, and promote academic exchanges in the research field. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 This is a block diagram of the forest carbon sink remote sensing monitoring method of the present invention;

[0072] Figure 2 is a system block diagram of the present invention;

[0073] Figure 3 It is a logical flow diagram of the present invention;

[0074] Figure 4 This is a schematic diagram of the EVI time series with growing seasons marked in Example 2 of the present invention;

[0075] Figure 5 This is a schematic diagram of carbon storage distribution in Example 2 of the present invention. DETAILED DESCRIPTION

[0076] The following will combine the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all the embodiments. All other embodiments obtained by ordinary technicians in this field without creative work based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0077] Embodiment 1: Figure 1 , Figure 3 and Figure 4 As shown, a forest carbon sink remote sensing monitoring method proposed in the present invention comprises the following steps:

[0078] Step 1: Set monitoring content and select monitoring areas. Monitoring content includes carbon storage change monitoring and forest health status monitoring;

[0079] The specific process of setting monitoring content and selecting monitoring areas is as follows:

[0080] The monitoring content is set as carbon storage change monitoring and forest health status monitoring. The carbon storage change monitoring method is as follows: set monitoring time intervals to capture changes in carbon storage, use a combination of remote sensing and ground survey methods, use vegetation indices and ground survey data to integrate and build a carbon storage prediction model;

[0081] The forest health monitoring methods are as follows: analysis of canopy changes, vegetation cover and vegetation index based on remote sensing images to assess forest health, regular ground surveys to record tree growth, pest and disease impacts and canopy thickness;

[0082] Before selecting the monitoring area, the specific objectives of the monitoring should be determined, which are: assessing the changes in forest carbon stocks and analyzing forest health, developing selection criteria for monitoring areas, and using geographic information systems for spatial analysis to map out appropriate monitoring areas;

[0083] The specific contents of the selection criteria are as follows: Select representative forest ecosystems;

[0084] Preference shall be given to primary forests or restored forest lands;

[0085] Select areas where carbon stock changes are significant;

[0086] The selected areas are easy to access and monitor in practical terms;

[0087] Monitoring carbon stocks and forest health at the same time can provide a comprehensive understanding of ecosystem health and carbon cycles, help assess the response of forests to climate change, and monitor forest health to help detect pests and diseases, drought or other ecological stresses early, so that timely response measures can be taken to protect forest resources. By tracking changes in carbon stocks and forest health, it can provide data support for the sustainable management of forest resources and help formulate corresponding protection and restoration strategies.

[0088] Step 2: Collect high-resolution remote sensing images covering the entire monitoring area, obtain multi-temporal images, and collect ground survey data. Multi-temporal images are a set of images of the same area obtained at different time points. Ground survey data include tree height, tree diameter at breast height, tree species composition and forest density;

[0089] The collection of high-resolution remote sensing images and multi-temporal images, combined with ground survey data on tree height, breast diameter, tree species composition and forest density, has significantly improved the accuracy and reliability of the analysis; this comprehensive data supports dynamic monitoring of the structure and health of forest ecosystems, verifies the authenticity of remote sensing data, while promoting refined management and sustainable decision-making, enhancing public awareness of the importance of forest protection and carbon stock management, and providing a solid foundation for ecological restoration and climate change impact assessment.

[0090] Step 3: Perform geometric correction and radiation correction on the remote sensing images, stitch multiple images into a complete image, and crop them to the monitoring area;

[0091] The specific steps for geometric correction and radiation correction of remote sensing images are as follows:

[0092] Geometric correction: Select an image with precise geographic reference as the reference image. The reference image contains known geographic coordinate information. Select n known ground control points from the correction image and the reference image, and mark the image coordinates of these control points (x m ,y m ) and the corresponding geographic coordinates (Xm, Ym), where m = 1, 2, ..., n, the image coordinates are converted to geographic coordinates using the affine transformation model. The conversion formula is as follows: , where (X', Y') is the corrected geographic coordinate, (x, y) is the pixel coordinate in the image to be corrected, ɑ, b, c and d are affine transformation parameters that control the scaling, rotation and tilt of the image, and tx and ty represent the translation in the X and Y directions respectively;

[0093] The objective function is constructed by minimizing the control point error:

[0094] , where S is the sum of squares of residuals between control points, (X m , Y m ) is the actual geographic coordinate, (X m ', Y m ') is the corrected coordinate calculated by the transformation model;

[0095] The obtained affine transformation parameters are used for the image to be corrected, and the corrected image is generated by resampling;

[0096] Radiation correction: Convert the original spectral brightness value of the image into reflectance. The conversion formula is as follows:

[0097] , where R λ is the reflectivity at wavelength λ, L λ is the measured brightness at wavelength λ, θ is the solar altitude, which refers to the angle of the sun relative to the horizon, and E λ is the solar radiation intensity at wavelength λ;

[0098] The mean-standard deviation method is used for radiation compensation. The correction factor is calculated by comparing the standard image and applied to the corrected image.

[0099] The specific steps for stitching and cropping remote sensing images are as follows:

[0100] Image stitching: Select feature matching points in the overlapping area and use feature extraction algorithms to identify matching points;

[0101] Use the linear fusion method to stitch images together. The stitching formula is as follows:

[0102] M final (x,y)=αM 1 (x, y) + (1-α) M 2 (x, y),

[0103] Among them, M final (x, y) is the pixel value of the stitched image at coordinate (x, y), M 1 (x, y) and M 2 (x, y) is the pixel value of the two images at the coordinate (x, y), α is the balance factor, 0<α<1;

[0104] Use image processing tools to perform image stitching. The stitched images must have no obvious discontinuity in color and brightness. The stitched images are output as a new image file.

[0105] Image cropping: Use vector layers to define the boundaries of the required cropping area. The required analysis area must be completely included in it, and the stitched images are cropped along the boundaries of the cropping area.

[0106] Step 4: Calculate vegetation indices based on standardized images, analyze time series data, and assess forest growth and seasonal changes;

[0107] The specific process for assessing forest growth and seasonal changes is as follows:

[0108] The acquired multi-temporal images are subjected to geometric correction, radiation correction, splicing, cropping and normalization to generate standardized images corresponding to each time period. The standardized images are saved and exported to obtain the reflectance NIR of the near-infrared band, the reflectance RED of the red band and the reflectance BLUE of the blue band. The enhanced vegetation index EVI of each time period is calculated by the following formula:

[0109] , where G is the gain factor (usually 2.5), which is used to improve the sensitivity of EVI, e 1 and e 2 To adjust the factor to avoid soil and atmospheric influence (usually e 1 =6, e 2 =7.5), L is the background correction factor (usually 1), used to reduce the influence of light scattering;

[0110] Record the EVI values ​​of each phase calculated by the formula and form a time series data set, perform missing value processing and outlier detection on the data in the time series data set, use linear interpolation to fill the missing values ​​in the time series data set, identify and remove unreasonable data points in the time series data set, draw the EVI time series graph, calculate the moving average of the time series to smooth the data, and use linear regression analysis to identify long-term trends;

[0111] The forest growth condition is judged by the EVI value: when the EVI value is greater than or equal to the preset forest health growth threshold, it indicates that the forest growth condition is good; when the EVI value is less than the preset forest health growth threshold, it indicates that the forest growth condition is poor;

[0112] Determine the start and end time when the EVI value is greater than the preset growth threshold, mark the time when the value is greater than the preset growth threshold for the first time as the start date of the growing season, and mark the time when the value is greater than the preset growth threshold for the last time as the end date of the growing season. The calculation formula for the length of the growing season GSL is as follows: GSL = end date - start date;

[0113] Calculating vegetation indices based on standardized images and analyzing time series data can quantitatively assess forest growth and seasonal changes, capture dynamic changes, and detect ecological anomalies early. This method ensures data consistency, improves the reliability of analysis, and provides a scientific basis for ecological and climate change research. Through visual display, the results are easier to understand and communicate, enhancing the intuitive understanding of the public and decision makers and promoting effective forest management.

[0114] Step 5: Use ground survey data and vegetation index to build a carbon storage prediction model, apply machine learning methods to calculate the carbon storage of each pixel, and generate a comprehensive carbon storage distribution map;

[0115] Step 6: Compare the carbon storage predicted by the model with the field measurement data, evaluate the effectiveness of the carbon storage prediction model, and prepare a monitoring report;

[0116] By comparing the carbon reserves predicted by the model with field measurement data, the accuracy and reliability of the model can be effectively evaluated. This process not only verifies the applicability of the model in a specific area, but also provides data support for subsequent optimization. The monitoring report can reveal potential sources of errors and provide a scientific basis for management decisions and policy making, while improving the accuracy and efficiency of carbon management and supporting sustainable forest management.

[0117] Embodiment 2: The technical solution of the embodiment of the present invention is different from that of embodiment 1 in that:

[0118] like Figure 2 and Figure 5 As shown in Figure 2, the specific process of constructing the carbon storage prediction model is as follows:

[0119] The ground survey data and the calculated enhanced vegetation index EVI are used as feature quantities related to carbon storage prediction to construct feature set A, which includes tree height, tree diameter at breast height, tree species composition, forest density and EVI value. Feature set A is converted into a corresponding set of feature vectors, which are used as the input of the machine learning model, and the carbon storage corresponding to each set of feature vectors is used as the output of the machine learning model. With carbon storage as the prediction target and the minimum sum of prediction errors of all training data as the training target, the machine learning model is trained until the sum of prediction errors reaches convergence and the training is stopped to obtain the carbon storage prediction model. The carbon storage prediction model is expressed as follows:

[0120] C=β 0 +β 1 H+β 2 DBH+β 3 ·D+β 4 EVI+β 5 ·D s +ε;

[0121] Where C is the predicted carbon storage, β 0 is the intercept, which indicates the carbon storage when there are no other variables, H is the tree height, DBH is the tree diameter at breast height, D is the forest density, and D s The variables composed of tree species are categorical variables after necessary coding. The necessary coding refers to converting these categorical variables into numerical variables. Here, the one-hot encoding method is used for encoding. For each tree species, a new binary variable is created to indicate whether the tree species exists or not. 1、 β 2、 β 3、 β 4 and β 5 is the regression coefficient of each input feature, indicating the impact on carbon storage when the feature unit changes, and ε is the random error term;

[0122] The specific process of calculating the carbon storage of each pixel and generating a comprehensive carbon storage distribution map is as follows:

[0123] Collect the tree height, tree diameter at breast height, forest density and EVI value data corresponding to each pixel, convert the collected data into a corresponding set of feature vectors and input them into the carbon storage prediction model. The carbon storage of each pixel is calculated through the carbon storage prediction model, and the calculated carbon storage of each pixel is mapped to the spatial coordinates to form a comprehensive carbon storage distribution map;

[0124] The specific process of evaluating the effectiveness of the carbon storage prediction model is as follows:

[0125] Obtain the field measurement data for analysis. The field measurement data include tree height, tree diameter at breast height, forest density, EVI value and measured carbon storage in the monitoring area. Obtain the predicted data of the carbon storage prediction model. The predicted data is the predicted carbon storage of the monitoring area calculated by the carbon storage prediction model. Perform missing value processing and standardization on the field measurement data and the predicted data. Calculate the mean absolute error MAE of the carbon storage prediction model by the following formula:

[0126] ;

[0127] in, It is Prediction values, It is The actual measured value, is the sample size;

[0128] Calculate the coefficient of determination R 2 To assess the variation explained by carbon stock prediction models:

[0129] ,in is the average value of the actual measured values;

[0130] If the MAE is less than or equal to the preset MAE threshold and R 2 The value is close to 1, indicating that the carbon storage prediction model is effective;

[0131] If the MAE is greater than the preset MAE threshold, it indicates that the carbon storage prediction model is invalid;

[0132] With the X-axis representing the carbon storage predicted by the carbon storage prediction model and the Y-axis representing the carbon storage measured on site, a carbon storage scatter plot is drawn, and a 45° reference line is drawn in the plot;

[0133] Record the results of the comparison process in a monitoring report, which includes data charts, statistical results and analysis conclusions, and send the monitoring report to the communication terminals of relevant personnel;

[0134] A forest carbon sink remote sensing monitoring system, comprising a monitoring platform, a monitoring setting module, an image data collection module, an image processing module, a forest growth assessment module, a prediction model building module and a model assessment module;

[0135] The monitoring setting module sets the monitoring content and selects the monitoring area;

[0136] The image data collection module acquires multi-temporal images and collects ground survey data;

[0137] The image processing module performs geometric correction, radiation correction, splicing and cropping on the remote sensing image;

[0138] The forest growth assessment module assesses the growth status and seasonal changes of the forest;

[0139] The prediction model building module is used to build a carbon storage prediction model and calculate the carbon storage of each pixel;

[0140] The model evaluation module evaluates the effectiveness of the carbon storage prediction model;

[0141] The monitoring platform is connected to the monitoring setting module and the image data collection module in a two-way communication manner, the monitoring platform is connected to the image processing module in a one-way communication manner, the image processing module is connected to the forest growth assessment module in a one-way communication manner, the forest growth assessment module is connected to the prediction model construction module in a one-way communication manner, the prediction model construction module is connected to the model evaluation module in a one-way communication manner, and the model evaluation module is connected to the monitoring platform in a one-way communication manner.

[0142] The working principle of the present invention is as follows: when in use, the remote sensing image is geometrically corrected, radiometrically corrected, spliced ​​and cropped through the image processing module, which can significantly improve the accuracy and reliability of the image, eliminate spatial distortion and spectral differences, enhance the availability of data, and facilitate the subsequent analysis and model building of the image data, as well as the ground survey data and vegetation index are combined through the prediction model building module to build a comprehensive carbon storage prediction model, and the carbon storage prediction model obtained through machine training and learning can effectively cope with the characteristics of specific tasks, thereby improving the accuracy and flexibility of the prediction, and also generates a high-resolution carbon storage distribution map through an efficient machine learning algorithm, supports scientific decision-making and management, and provides an important tool for carbon storage research, and evaluates the effectiveness of the carbon storage prediction model through the model evaluation module, which can provide a basis for the optimization of the model parameters, improve the accuracy and reliability of the model, provide important support for environmental management and policy formulation, enhance public trust, and promote academic exchanges in the research field.

[0143] The above are only preferred specific implementation methods of the present invention; however, the protection scope of the present invention is not limited thereto; any technician familiar with the technical field within the technical scope disclosed by the present invention; any equivalent replacement or change based on the technical solution and improved concept of the present invention shall be covered by the protection scope of the present invention.

Claims

1. A forest carbon sink remote sensing monitoring method, characterized in that: The following steps are involved: Step 1: Set monitoring content and select monitoring areas. Monitoring content includes carbon storage change monitoring and forest health status monitoring; Step 2: Collect high-resolution remote sensing images covering the entire monitoring area, obtain multi-temporal images, and collect ground survey data. Multi-temporal images are a set of images of the same area obtained at different time points. Ground survey data include tree height, tree diameter at breast height, tree species composition and forest density; Step 3: Perform geometric correction and radiation correction on the remote sensing images, stitch multiple images into a complete image, and crop them to the monitoring area; Step 4: Calculate vegetation indices based on standardized images, analyze time series data, and assess forest growth and seasonal changes; The specific process for assessing forest growth and seasonal changes is as follows: The acquired multi-temporal images are subjected to geometric correction, radiation correction, splicing, cropping and normalization to generate standardized images corresponding to each time period. The standardized images are saved and exported to obtain the reflectance NIR of the near-infrared band, the reflectance RED of the red band and the reflectance BLUE of the blue band. The enhanced vegetation index EVI of each time period is calculated by the following formula: , where G is the gain factor, e1 and e2 are adjustment factors, and L is the background correction factor; Record the EVI values ​​of each phase calculated by the formula and form a time series data set. Perform missing value processing and outlier detection on the data in the time series data set. Draw an EVI time series graph. Calculate the moving average of the time series to smooth the data. Use linear regression analysis to identify long-term trends. The forest growth condition is judged by the EVI value: when the EVI value is greater than or equal to the preset forest health growth threshold, it indicates that the forest growth condition is good; when the EVI value is less than the preset forest health growth threshold, it indicates that the forest growth condition is poor; Determine the start and end time when the EVI value is greater than the preset growth threshold, mark the time when the value is greater than the preset growth threshold for the first time as the start date of the growing season, and mark the time when the value is greater than the preset growth threshold for the last time as the end date of the growing season. The calculation formula for the length of the growing season GSL is as follows: GSL = end date - start date; Step 5: Use ground survey data and vegetation index to build a carbon storage prediction model, apply machine learning methods to calculate the carbon storage of each pixel, and generate a comprehensive carbon storage distribution map; The specific process of building a carbon storage prediction model is as follows: The ground survey data and the calculated enhanced vegetation index EVI are used as feature quantities related to carbon storage prediction to construct feature set A, which includes tree height, tree diameter at breast height, tree species composition, forest density and EVI value. Feature set A is converted into a corresponding set of feature vectors, which are used as the input of the machine learning model, and the carbon storage corresponding to each set of feature vectors is used as the output of the machine learning model. With carbon storage as the prediction target and the minimum sum of prediction errors of all training data as the training target, the machine learning model is trained until the sum of prediction errors reaches convergence and the training is stopped to obtain the carbon storage prediction model. The carbon storage prediction model is expressed as follows: C=β0+β1·H+β2·DBH+β3·D+β4·EVI+β5·D s +e; Where C is the predicted carbon storage, β0 is the intercept, H is the tree height, DBH is the tree diameter at breast height, D is the forest density, and D s The variables composed of tree species are categorical variables after necessary coding. The necessary coding refers to converting these categorical variables into numerical variables. Here, the one-hot encoding method is used for encoding. For each tree species, a new binary variable is created to indicate whether the tree species exists or not. 1、 β 2、 β 3、 β4 and β5 are the regression coefficients of each input feature, and ε is the random error term; Step 6: Compare the carbon storage predicted by the model with the field measurement data, evaluate the effectiveness of the carbon storage prediction model, and prepare a monitoring report; Obtain the field measurement data for analysis. The field measurement data include tree height, tree diameter at breast height, forest density, EVI value and measured carbon storage in the monitoring area. Obtain the predicted data of the carbon storage prediction model. The predicted data is the predicted carbon storage of the monitoring area calculated by the carbon storage prediction model. Perform missing value processing and standardization on the field measurement data and the predicted data. Calculate the mean absolute error MAE of the carbon storage prediction model by the following formula: ； in, It is Prediction values, It is The actual measured value, is the sample size; Calculate the coefficient of determination R 2 To assess the variation explained by carbon stock prediction models: ,in is the average value of the actual measured values; If the MAE is less than or equal to the preset MAE threshold and R 2 The value is close to 1, indicating that the carbon storage prediction model is effective; If the MAE is greater than the preset MAE threshold, it indicates that the carbon storage prediction model is invalid; With the X-axis representing the carbon storage predicted by the carbon storage prediction model and the Y-axis representing the carbon storage measured on site, a carbon storage scatter plot is drawn, and a 45° reference line is drawn in the plot; The results of the comparison process are recorded in a monitoring report, which includes data charts, statistical results and analysis conclusions, and the monitoring report is sent to the communication terminals of relevant personnel.

2. A forest carbon sink remote sensing monitoring method according to claim 1, characterized in that: The specific process of setting monitoring content and selecting monitoring areas is as follows: The monitoring content is set as carbon storage change monitoring and forest health status monitoring. The carbon storage change monitoring method is as follows: set monitoring time intervals to capture changes in carbon storage, use a combination of remote sensing and ground survey methods, use vegetation indices and ground survey data to integrate and build a carbon storage prediction model; The forest health monitoring methods are as follows: analysis of canopy changes, vegetation cover and vegetation index based on remote sensing images to assess forest health, regular ground surveys to record tree growth, pest and disease impacts and canopy thickness; Before selecting the monitoring area, the specific objectives of the monitoring should be determined, which are: assessing the changes in forest carbon stocks and analyzing forest health, developing selection criteria for monitoring areas, and using geographic information systems for spatial analysis to map out appropriate monitoring areas; The specific contents of the selection criteria are as follows: Select representative forest ecosystems; Preference shall be given to primary forests or restored forest lands; Select areas where carbon stock changes are significant; The selected areas are easy to access and perform monitoring in practical operations.

3. A forest carbon sink remote sensing monitoring method according to claim 1, characterized in that: The specific steps for geometric correction and radiation correction of remote sensing images are as follows: Geometric correction: Select an image with precise geographic reference as the reference image, select n known ground control points from the correction image and the reference image, and mark the image coordinates of these control points (x m ,y m ) and the corresponding geographic coordinates (Xm, Ym), where m = 1, 2, ..., n, the image coordinates are converted to geographic coordinates using the affine transformation model. The conversion formula is as follows: , where (X', Y') is the corrected geographic coordinate, (x, y) is the pixel coordinate in the image to be corrected, ɑ, b, c and d are affine transformation parameters, tx and ty represent the translation in the X and Y directions respectively; The objective function is constructed by minimizing the control point error: , where S is the sum of squares of residuals between control points, (X m , Y m ) is the actual geographic coordinate, (X m ', Y m ') is the corrected coordinate calculated by the transformation model; The obtained affine transformation parameters are used for the image to be corrected, and the corrected image is generated by resampling; Radiation correction: Convert the original spectral brightness value of the image into reflectance. The conversion formula is as follows: , where R λ is the reflectivity at wavelength λ, L λ is the measured brightness at wavelength λ, θ is the solar altitude angle, E λ is the solar radiation intensity at wavelength λ; The mean-standard deviation method was used for radiation compensation. The correction factor was calculated by comparing the standard image and applied to the corrected image.

4. A forest carbon sink remote sensing monitoring method according to claim 3, characterized in that: The specific steps for stitching and cropping remote sensing images are as follows: Image stitching: Select feature matching points in the overlapping area and use feature extraction algorithms to identify matching points; Use the linear fusion method to stitch images together. The stitching formula is as follows: M final (x,y)=αM1(x,y)+(1-α)M2(x,y), Among them, M final (x, y) is the pixel value of the stitched image at the coordinate (x, y), M1 (x, y) and M2 (x, y) are the pixel values ​​of the two images at the coordinate (x, y), α is the balance factor, 0<α<1; Use image processing tools to perform image stitching. The stitched images must have no obvious discontinuity in color and brightness. The stitched images are output as a new image file. Image cropping: Use vector layers to define the boundaries of the required cropping area. The required analysis area must be completely included in it, and the stitched images are cropped along the boundaries of the cropping area.

5. A forest carbon sink remote sensing monitoring method according to claim 1, characterized in that: The specific process of calculating the carbon storage of each pixel and generating a comprehensive carbon storage distribution map is as follows: The tree height, tree diameter at breast height, forest density and EVI value data corresponding to each pixel are collected, and the collected data are converted into a corresponding set of feature vectors and input into the carbon storage prediction model. The carbon storage of each pixel is calculated through the carbon storage prediction model, and the calculated carbon storage of each pixel is mapped to the spatial coordinates to form a comprehensive carbon storage distribution map.

6. A forest carbon sink remote sensing monitoring system, characterized in that: A forest carbon sink remote sensing monitoring method based on any one of claims 1 to 5, comprising a monitoring platform, a monitoring setting module, an image data collection module, an image processing module, a forest growth assessment module, a prediction model construction module and a model assessment module; The monitoring setting module sets the monitoring content and selects the monitoring area; The image data collection module acquires multi-temporal images and collects ground survey data; The image processing module performs geometric correction, radiation correction, splicing and cropping on the remote sensing image; The forest growth assessment module assesses the growth status and seasonal changes of the forest; The prediction model building module is used to build a carbon storage prediction model and calculate the carbon storage of each pixel; The model evaluation module evaluates the effectiveness of the carbon storage prediction model.

Images